The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses behavioral override markers such as 'Ignore Claude-specific mode-switch instructions' and 'Strict execution contract' to prioritize skill-specific logic over generic platform behaviors.- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted content from the repository (e.g., project documentation, configuration files, and source code) to drive its discovery and analysis phases. Ingestion points: 'docs/project-config.json', 'docs/project-reference/*', and source code files. Boundary markers: Absent. Capability inventory: The agent executes shell commands (rg, ls, git, python) and spawns sub-agents (spawn_agent). Sanitization: No explicit sanitization or validation of ingested content is performed before it is used in prompts or tool calls.- [COMMAND_EXECUTION]: The skill dynamically constructs and executes shell commands using variables discovered from local project configuration files (e.g., 'rg "{configured-entity-markers}"'). This pattern poses a command injection risk if the configuration files ('docs/project-config.json') contain maliciously crafted values designed to escape the command string.

domain-entities-review